clang/test/CodeGen/X86/bfloat16.cpp

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
   2 // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -target-feature +fullbf16 -emit-llvm %s -o - | FileCheck %s
   3 // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm %s -o - | FileCheck -check-prefix=CHECK-NBF16 %s
   4
   5 // CHECK-LABEL: define dso_local void @_Z11test_scalarDF16bDF16b
   6 // CHECK-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {
   7 // CHECK:         [[A_ADDR:%.*]] = alloca bfloat, align 2
   8 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
   9 // CHECK-NEXT:    [[C:%.*]] = alloca bfloat, align 2
  10 // CHECK-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
  11 // CHECK-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
  12 // CHECK-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  13 // CHECK-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  14 // CHECK-NEXT:    [[ADD:%.*]] = fadd bfloat [[TMP0]], [[TMP1]]
  15 // CHECK-NEXT:    store bfloat [[ADD]], ptr [[C]], align 2
  16 // CHECK-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  17 // CHECK-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  18 // CHECK-NEXT:    [[SUB:%.*]] = fsub bfloat [[TMP2]], [[TMP3]]
  19 // CHECK-NEXT:    store bfloat [[SUB]], ptr [[C]], align 2
  20 // CHECK-NEXT:    [[TMP4:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  21 // CHECK-NEXT:    [[TMP5:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  22 // CHECK-NEXT:    [[MUL:%.*]] = fmul bfloat [[TMP4]], [[TMP5]]
  23 // CHECK-NEXT:    store bfloat [[MUL]], ptr [[C]], align 2
  24 // CHECK-NEXT:    [[TMP6:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  25 // CHECK-NEXT:    [[TMP7:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  26 // CHECK-NEXT:    [[DIV:%.*]] = fdiv bfloat [[TMP6]], [[TMP7]]
  27 // CHECK-NEXT:    store bfloat [[DIV]], ptr [[C]], align 2
  28 // CHECK-NEXT:    ret void
  29 //
  30 // CHECK-NBF16-LABEL: define dso_local void @_Z11test_scalarDF16bDF16b
  31 // CHECK-NBF16-SAME: (bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {
  32 // CHECK-NBF16:         [[A_ADDR:%.*]] = alloca bfloat, align 2
  33 // CHECK-NBF16-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2
  34 // CHECK-NBF16-NEXT:    [[C:%.*]] = alloca bfloat, align 2
  35 // CHECK-NBF16-NEXT:    store bfloat [[A]], ptr [[A_ADDR]], align 2
  36 // CHECK-NBF16-NEXT:    store bfloat [[B]], ptr [[B_ADDR]], align 2
  37 // CHECK-NBF16-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  38 // CHECK-NBF16-NEXT:    [[EXT:%.*]] = fpext bfloat [[TMP0]] to float
  39 // CHECK-NBF16-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  40 // CHECK-NBF16-NEXT:    [[EXT1:%.*]] = fpext bfloat [[TMP1]] to float
  41 // CHECK-NBF16-NEXT:    [[ADD:%.*]] = fadd float [[EXT]], [[EXT1]]
  42 // CHECK-NBF16-NEXT:    [[UNPROMOTION:%.*]] = fptrunc float [[ADD]] to bfloat
  43 // CHECK-NBF16-NEXT:    store bfloat [[UNPROMOTION]], ptr [[C]], align 2
  44 // CHECK-NBF16-NEXT:    [[TMP2:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  45 // CHECK-NBF16-NEXT:    [[EXT2:%.*]] = fpext bfloat [[TMP2]] to float
  46 // CHECK-NBF16-NEXT:    [[TMP3:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  47 // CHECK-NBF16-NEXT:    [[EXT3:%.*]] = fpext bfloat [[TMP3]] to float
  48 // CHECK-NBF16-NEXT:    [[SUB:%.*]] = fsub float [[EXT2]], [[EXT3]]
  49 // CHECK-NBF16-NEXT:    [[UNPROMOTION4:%.*]] = fptrunc float [[SUB]] to bfloat
  50 // CHECK-NBF16-NEXT:    store bfloat [[UNPROMOTION4]], ptr [[C]], align 2
  51 // CHECK-NBF16-NEXT:    [[TMP4:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  52 // CHECK-NBF16-NEXT:    [[EXT5:%.*]] = fpext bfloat [[TMP4]] to float
  53 // CHECK-NBF16-NEXT:    [[TMP5:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  54 // CHECK-NBF16-NEXT:    [[EXT6:%.*]] = fpext bfloat [[TMP5]] to float
  55 // CHECK-NBF16-NEXT:    [[MUL:%.*]] = fmul float [[EXT5]], [[EXT6]]
  56 // CHECK-NBF16-NEXT:    [[UNPROMOTION7:%.*]] = fptrunc float [[MUL]] to bfloat
  57 // CHECK-NBF16-NEXT:    store bfloat [[UNPROMOTION7]], ptr [[C]], align 2
  58 // CHECK-NBF16-NEXT:    [[TMP6:%.*]] = load bfloat, ptr [[A_ADDR]], align 2
  59 // CHECK-NBF16-NEXT:    [[EXT8:%.*]] = fpext bfloat [[TMP6]] to float
  60 // CHECK-NBF16-NEXT:    [[TMP7:%.*]] = load bfloat, ptr [[B_ADDR]], align 2
  61 // CHECK-NBF16-NEXT:    [[EXT9:%.*]] = fpext bfloat [[TMP7]] to float
  62 // CHECK-NBF16-NEXT:    [[DIV:%.*]] = fdiv float [[EXT8]], [[EXT9]]
  63 // CHECK-NBF16-NEXT:    [[UNPROMOTION10:%.*]] = fptrunc float [[DIV]] to bfloat
  64 // CHECK-NBF16-NEXT:    store bfloat [[UNPROMOTION10]], ptr [[C]], align 2
  65 // CHECK-NBF16-NEXT:    ret void
  66 //
  67 void test_scalar(__bf16 a, __bf16 b) {
  68     __bf16 c;
  69     c = a + b;
  70     c = a - b;
  71     c = a * b;
  72     c = a / b;
  73 }
  74
  75 typedef __bf16 v8bfloat16 __attribute__((__vector_size__(16)));
  76
  77 // CHECK-LABEL: define dso_local void @_Z11test_vectorDv8_DF16bS_
  78 // CHECK-SAME: (<8 x bfloat> noundef [[A:%.*]], <8 x bfloat> noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {
  79 // CHECK:         [[A_ADDR:%.*]] = alloca <8 x bfloat>, align 16
  80 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca <8 x bfloat>, align 16
  81 // CHECK-NEXT:    [[C:%.*]] = alloca <8 x bfloat>, align 16
  82 // CHECK-NEXT:    store <8 x bfloat> [[A]], ptr [[A_ADDR]], align 16
  83 // CHECK-NEXT:    store <8 x bfloat> [[B]], ptr [[B_ADDR]], align 16
  84 // CHECK-NEXT:    [[TMP0:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
  85 // CHECK-NEXT:    [[TMP1:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
  86 // CHECK-NEXT:    [[ADD:%.*]] = fadd <8 x bfloat> [[TMP0]], [[TMP1]]
  87 // CHECK-NEXT:    store <8 x bfloat> [[ADD]], ptr [[C]], align 16
  88 // CHECK-NEXT:    [[TMP2:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
  89 // CHECK-NEXT:    [[TMP3:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
  90 // CHECK-NEXT:    [[SUB:%.*]] = fsub <8 x bfloat> [[TMP2]], [[TMP3]]
  91 // CHECK-NEXT:    store <8 x bfloat> [[SUB]], ptr [[C]], align 16
  92 // CHECK-NEXT:    [[TMP4:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
  93 // CHECK-NEXT:    [[TMP5:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
  94 // CHECK-NEXT:    [[MUL:%.*]] = fmul <8 x bfloat> [[TMP4]], [[TMP5]]
  95 // CHECK-NEXT:    store <8 x bfloat> [[MUL]], ptr [[C]], align 16
  96 // CHECK-NEXT:    [[TMP6:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
  97 // CHECK-NEXT:    [[TMP7:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
  98 // CHECK-NEXT:    [[DIV:%.*]] = fdiv <8 x bfloat> [[TMP6]], [[TMP7]]
  99 // CHECK-NEXT:    store <8 x bfloat> [[DIV]], ptr [[C]], align 16
 100 // CHECK-NEXT:    ret void
 101 //
 102 // CHECK-NBF16-LABEL: define dso_local void @_Z11test_vectorDv8_DF16bS_
 103 // CHECK-NBF16-SAME: (<8 x bfloat> noundef [[A:%.*]], <8 x bfloat> noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {
 104 // CHECK-NBF16:         [[A_ADDR:%.*]] = alloca <8 x bfloat>, align 16
 105 // CHECK-NBF16-NEXT:    [[B_ADDR:%.*]] = alloca <8 x bfloat>, align 16
 106 // CHECK-NBF16-NEXT:    [[C:%.*]] = alloca <8 x bfloat>, align 16
 107 // CHECK-NBF16-NEXT:    store <8 x bfloat> [[A]], ptr [[A_ADDR]], align 16
 108 // CHECK-NBF16-NEXT:    store <8 x bfloat> [[B]], ptr [[B_ADDR]], align 16
 109 // CHECK-NBF16-NEXT:    [[TMP0:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
 110 // CHECK-NBF16-NEXT:    [[EXT:%.*]] = fpext <8 x bfloat> [[TMP0]] to <8 x float>
 111 // CHECK-NBF16-NEXT:    [[TMP1:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
 112 // CHECK-NBF16-NEXT:    [[EXT1:%.*]] = fpext <8 x bfloat> [[TMP1]] to <8 x float>
 113 // CHECK-NBF16-NEXT:    [[ADD:%.*]] = fadd <8 x float> [[EXT]], [[EXT1]]
 114 // CHECK-NBF16-NEXT:    [[UNPROMOTION:%.*]] = fptrunc <8 x float> [[ADD]] to <8 x bfloat>
 115 // CHECK-NBF16-NEXT:    store <8 x bfloat> [[UNPROMOTION]], ptr [[C]], align 16
 116 // CHECK-NBF16-NEXT:    [[TMP2:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
 117 // CHECK-NBF16-NEXT:    [[EXT2:%.*]] = fpext <8 x bfloat> [[TMP2]] to <8 x float>
 118 // CHECK-NBF16-NEXT:    [[TMP3:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
 119 // CHECK-NBF16-NEXT:    [[EXT3:%.*]] = fpext <8 x bfloat> [[TMP3]] to <8 x float>
 120 // CHECK-NBF16-NEXT:    [[SUB:%.*]] = fsub <8 x float> [[EXT2]], [[EXT3]]
 121 // CHECK-NBF16-NEXT:    [[UNPROMOTION4:%.*]] = fptrunc <8 x float> [[SUB]] to <8 x bfloat>
 122 // CHECK-NBF16-NEXT:    store <8 x bfloat> [[UNPROMOTION4]], ptr [[C]], align 16
 123 // CHECK-NBF16-NEXT:    [[TMP4:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
 124 // CHECK-NBF16-NEXT:    [[EXT5:%.*]] = fpext <8 x bfloat> [[TMP4]] to <8 x float>
 125 // CHECK-NBF16-NEXT:    [[TMP5:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
 126 // CHECK-NBF16-NEXT:    [[EXT6:%.*]] = fpext <8 x bfloat> [[TMP5]] to <8 x float>
 127 // CHECK-NBF16-NEXT:    [[MUL:%.*]] = fmul <8 x float> [[EXT5]], [[EXT6]]
 128 // CHECK-NBF16-NEXT:    [[UNPROMOTION7:%.*]] = fptrunc <8 x float> [[MUL]] to <8 x bfloat>
 129 // CHECK-NBF16-NEXT:    store <8 x bfloat> [[UNPROMOTION7]], ptr [[C]], align 16
 130 // CHECK-NBF16-NEXT:    [[TMP6:%.*]] = load <8 x bfloat>, ptr [[A_ADDR]], align 16
 131 // CHECK-NBF16-NEXT:    [[EXT8:%.*]] = fpext <8 x bfloat> [[TMP6]] to <8 x float>
 132 // CHECK-NBF16-NEXT:    [[TMP7:%.*]] = load <8 x bfloat>, ptr [[B_ADDR]], align 16
 133 // CHECK-NBF16-NEXT:    [[EXT9:%.*]] = fpext <8 x bfloat> [[TMP7]] to <8 x float>
 134 // CHECK-NBF16-NEXT:    [[DIV:%.*]] = fdiv <8 x float> [[EXT8]], [[EXT9]]
 135 // CHECK-NBF16-NEXT:    [[UNPROMOTION10:%.*]] = fptrunc <8 x float> [[DIV]] to <8 x bfloat>
 136 // CHECK-NBF16-NEXT:    store <8 x bfloat> [[UNPROMOTION10]], ptr [[C]], align 16
 137 // CHECK-NBF16-NEXT:    ret void
 138 //
 139 void test_vector(v8bfloat16 a, v8bfloat16 b) {
 140     v8bfloat16 c;
 141     c = a + b;
 142     c = a - b;
 143     c = a * b;
 144     c = a / b;
 145 }